talk-llama : sync llama.cpp

This commit is contained in:
Georgi Gerganov
2024-01-14 00:13:17 +02:00
parent c615f2c335
commit f001a3b7b6
2 changed files with 72 additions and 55 deletions

View File

@@ -43,7 +43,7 @@
#define LLAMA_FILE_MAGIC_GGSN 0x6767736eu // 'ggsn'
#define LLAMA_SESSION_MAGIC LLAMA_FILE_MAGIC_GGSN
#define LLAMA_SESSION_VERSION 3
#define LLAMA_SESSION_VERSION 4
#if defined(GGML_USE_CUBLAS) || defined(GGML_USE_CLBLAST) || defined(GGML_USE_METAL)
// Defined when llama.cpp is compiled with support for offloading model layers to GPU.